What is a prompt?
The dictionary defines the verb “to prompt” as: to make something happen; or, to help someone (such as an actor) remember what they were going to say (which I take to mean as a tool for stimulating memory); or, finally, to give an instruction to AI (artificial intelligence) in natural language rather than in computer language (code).
What prompts are NOT are search parameters. We are not prompting when we ask Google for an image of a red school bus with a flat tire on an icy road in winter. Those are search terms and there is a finite, though often huge number of responses that meet all of the criteria. A prompt, on the other hand, is something more emotional, more creative. And although it can feel like the two are the same, they are most decidedly two different entities. The two processes result in decidedly different results, as well. A search will yield the match to the criteria, either exact or approximate. The prompt will yield something the algorithm interprets as a visual image of the author’s description/instruction.
I’ve been co-leading a workshop on Generative AI and the Creative Process in a class for Maine Media College (where I am doing my MFA). One of the most important takeaways from the class, at least for me in my artistic process, is the value of learning how to prompt. How to prompt well, that is.
Prompts that work well are more like poetry than like an operator’s manual. Strong prompts suggest an environment. They give a sense of time of day, lighting, mood, weather, geography, decade, air quality, ambient noise, the feeling of the heat, the chill of the cold, the music in the background, the smell in the air.
The emotional qualities of our prompts affect the resulting images in AI. And this has been tested and born out with research. With a few well-placed, carefully chosen emotional phrases, we can garner AI responses (images) attaining heightened depth and “correctness” (matching our intention for the prompt).
The use of moderate emotional language (refrain from the purple prose) appears to push the generative AI to be more strident in generating a response. Having said that, let’s not make the mistake of believing this means AI is sentient. IT IS NOT. Repeat: IT IS NOT. But this is true: Generative AI has been data-trained on human writing and is therefore sensitive to emotionally-laden language and responses. Generative AI is a large-scale computational pattern-matching mimicry consisting of what we humans would construe as “understanding” and “knowledge.”
Put simply, the AI algorithms are already good enough to mathematically gauge when emotional language comes into play. And to respond accordingly! To give the image that extra “oomph,” if you will. Or, as I call it, that storytelling quality.
Consider the story-making possibilities!
Here’s a prompt:
“A chilly, raw February day in downtown Chicago, the Windy City, at mid-day in 1948. The sky is already turning dark and there is the unmistakable smell of snow in the air. A 40-year old man stands at the street corner sucking in the last drag of his cigarette and getting ready to drop the stub on the sidewalk. He is pre-occupied and running late. He has been waiting for a bus and the bus is late, probably because of the bad traffic. He’s worried he might be fired if he is late again.”
And here are two results.
I could start a story from these images!
Now let’s think about the implications for creative practice.
For the image at the top of this post, I was given an audio prompt by my mentor: the unmistakable sound of an aircraft passing overhead. Within seconds — literally — I was transported by memory to being nine years old and playing in the backyard of my parents’ suburban home in West Hartford, Connecticut. Fifty six years later, I still remember the sound of that plane, and the sight of its monstrously large form overhead. It was flying very low. Way, way lower than any plane had ever flown over our neighborhood before. And it scared me. And I remember it.
The prompt was designed to do exactly what prompts are, by definition, intended to do:to make something happen; or, to help someone (such as an actor) remember what they were going to say (which I take to mean as a tool for stimulating memory); or, finally, to give an instruction to AI (artificial intelligence) in natural language rather than in computer language (code).
My memory was jogged. A vision came into my mind. I gave Chat GPT 4 (the prompting mechanism for DALL-E 3) a brief description and this is what came out.
Here is the prompt I used:
“Create an image of a 9-year old girl in her suburban backyard in 1968. A huge plane is flying low. She thinks the Russians are coming. [A popular fear in the Cold War days of my childhood.] Black and white cinematic style. Film noir.”
And voila!
What’s important here is not the image itself. In fact, to be honest, the image is beside the point although I kind of like it. What matters is that this one prompt and this resulting image has now brought to mind the memory of the low-flying plane. The fear of invasion by “enemy forces.” The memory of the backyard swing set. The vulnerability of being a child outside the house while my mother and father were inside the house. Standing alone. Frightened. Small. Innocent. And carrying the fears of the nation.
Think about how much of my life and my thinking was influenced by this sense of vulnerability, and this sense of being a citizen of “the good guys’ nation.” Hmmmm.
This treasure trove of associations springs from a single prompt.
This is what imagination looks like. And feels like. We ALL have this. Whether we use AI to help generate ideas or use it to create visual work or prefer pen and ink or collage or a good game of charades, for that matter, it’s about the limitless capacity for curiosity and imagination that counts.
My thoughts on a dreary Sunday in early April, not a swing set in sight.